[flang][OpenMP] Delayed privatization MLIR lowering support for distribute

Starts delayed privatizaiton support for standalone `distribute`
directives. Other flavours of `distribute` are still TODO as well as
MLIR to LLVM IR lowering.

Author: ergawy

flang/lib/Lower/OpenMP/OpenMP.cpp

@@ -1681,7 +1681,6 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
                    mapTypes, deviceAddrSyms, deviceAddrLocs, deviceAddrTypes,
                    devicePtrSyms, devicePtrLocs, devicePtrTypes);
 
-  llvm::SmallVector<const semantics::Symbol *> privateSyms;
   DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
                            /*shouldCollectPreDeterminedSymbols=*/
                            lower::omp::isLastItemInQueue(item, queue),
@@ -1936,24 +1935,35 @@ static void genStandaloneDistribute(lower::AbstractConverter &converter,
   genDistributeClauses(converter, semaCtx, stmtCtx, item->clauses, loc,
                        distributeClauseOps);
 
-  // TODO: Support delayed privatization.
   DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
                            /*shouldCollectPreDeterminedSymbols=*/true,
-                           /*useDelayedPrivatization=*/false, &symTable);
-  dsp.processStep1();
+                           enableDelayedPrivatizationStaging, &symTable);
+  mlir::omp::PrivateClauseOps privateClauseOps;
+  dsp.processStep1(&privateClauseOps);
+  llvm::SmallVector<mlir::Type> privateVarTypes{};
+
+  for (mlir::Value privateVar : privateClauseOps.privateVars)
+    privateVarTypes.push_back(privateVar.getType());
 
   mlir::omp::LoopNestOperands loopNestClauseOps;
   llvm::SmallVector<const semantics::Symbol *> iv;
   genLoopNestClauses(converter, semaCtx, eval, item->clauses, loc,
                      loopNestClauseOps, iv);
 
-  // TODO: Populate entry block arguments with private variables.
   auto distributeOp = genWrapperOp<mlir::omp::DistributeOp>(
-      converter, loc, distributeClauseOps, /*blockArgTypes=*/{});
+      converter, loc, distributeClauseOps, privateVarTypes);
+  distributeOp.setPrivateSymsAttr(
+      privateClauseOps.privateSyms.empty()
+          ? nullptr
+          : converter.getFirOpBuilder().getArrayAttr(
+                privateClauseOps.privateSyms));
+
+  for (mlir::Value privateVar : privateClauseOps.privateVars)
+    distributeOp.getPrivateVarsMutable().append(privateVar);
 
   genLoopNestOp(converter, symTable, semaCtx, eval, loc, queue, item,
-                loopNestClauseOps, iv,
-                /*wrapperSyms=*/{}, distributeOp.getRegion().getArguments(),
+                loopNestClauseOps, iv, dsp.getDelayedPrivSymbols(),
+                distributeOp.getRegion().getArguments(),
                 llvm::omp::Directive::OMPD_distribute, dsp);
 }
 

flang/test/Lower/OpenMP/DelayedPrivatization/distribute-standalone-private.f90

@@ -0,0 +1,42 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -mmlir --openmp-enable-delayed-privatization-staging \
+! RUN:   -o - %s 2>&1 | FileCheck %s
+! RUN: bbc -emit-hlfir -fopenmp --openmp-enable-delayed-privatization-staging -o - %s 2>&1 \
+! RUN:   | FileCheck %s
+
+subroutine standalone_distribute
+    implicit none
+    integer :: simple_var, i
+
+    !$omp teams
+    !$omp distribute private(simple_var)
+    do i = 1, 10
+      simple_var = simple_var + i
+    end do
+    !$omp end distribute
+    !$omp end teams
+end subroutine standalone_distribute
+
+! CHECK: omp.private {type = private} @[[I_PRIVATIZER_SYM:.*]] : !fir.ref<i32>
+! CHECK: omp.private {type = private} @[[VAR_PRIVATIZER_SYM:.*]] : !fir.ref<i32>
+
+
+! CHECK-LABEL: func.func @_QPstandalone_distribute() {
+! CHECK:         %[[I_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFstandalone_distributeEi"}
+! CHECK:         %[[VAR_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFstandalone_distributeEsimple_var"}
+! CHECK:         omp.teams {
+! CHECK:           omp.distribute
+! CHECK-SAME:        private(@[[VAR_PRIVATIZER_SYM]] %[[VAR_DECL]]#0 -> %[[VAR_ARG:.*]] : !fir.ref<i32>,
+! CHECK-SAME:                @[[I_PRIVATIZER_SYM]] %[[I_DECL]]#0 -> %[[I_ARG:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.loop_nest {{.*}} {
+! CHECK:               %[[VAR_PRIV_DECL:.*]]:2 = hlfir.declare %[[VAR_ARG]]
+! CHECK:               %[[I_PRIV_DECL:.*]]:2 = hlfir.declare %[[I_ARG]]
+
+! CHECK:               fir.store %{{.*}} to %[[I_PRIV_DECL]]#1 : !fir.ref<i32>
+! CHECK:               %{{.*}} = fir.load %[[VAR_PRIV_DECL]]#0 : !fir.ref<i32>
+! CHECK:               %{{.*}} = fir.load %[[I_PRIV_DECL]]#0 : !fir.ref<i32>
+! CHECK:               arith.addi %{{.*}}, %{{.*}} : i32
+! CHECK:               hlfir.assign %{{.*}} to %[[VAR_PRIV_DECL]]#0 : i32, !fir.ref<i32>
+! CHECK:             }
+! CHECK:           }
+! CHECK:         }
+! CHECK:       }

mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp

@@ -1225,17 +1225,13 @@ parsePrivateList(OpAsmParser &parser,
 }
 
 static void printPrivateList(OpAsmPrinter &p, Operation *op,
-                             ValueRange privateVars, TypeRange privateTypes,
-                             ArrayAttr privateSyms) {
-  // TODO: Remove target-specific logic from this function.
-  auto targetOp = mlir::dyn_cast<mlir::omp::TargetOp>(op);
-  assert(targetOp);
-
+                             Operation::operand_range privateVars,
+                             TypeRange privateTypes, ArrayAttr privateSyms) {
   auto &region = op->getRegion(0);
   auto *argsBegin = region.front().getArguments().begin();
-  MutableArrayRef argsSubrange(argsBegin + targetOp.getMapVars().size(),
-                               argsBegin + targetOp.getMapVars().size() +
-                                   privateTypes.size());
+  MutableArrayRef argsSubrange(argsBegin + privateVars.getBeginOperandIndex(),
+                               argsBegin + privateVars.getBeginOperandIndex() +
+                                   privateVars.size());
   mlir::SmallVector<bool> isByRefVec;
   isByRefVec.resize(privateTypes.size(), false);
   DenseBoolArrayAttr isByRef =